How We Calculate the Start Score (And Why We Show the Math)

Fantasy football rankings are a black box. You've seen the sites: a 200-player board sorted by some composite number, no explanation of how it was built, no indication of which inputs moved it, no way to check the work. You're supposed to just trust it.

The Start Score takes the opposite position. Every component of every rating is documented here. If you think we weighted something wrong, you can make that argument with specifics. That's not a disclaimer — it's the point.

The formula, plain and simple

The Start Score projects each player's PPR fantasy points for the upcoming week, then converts that projection to a 0–100 percentile rank within the position.

The projection is: Role × Efficiency × Matchup × Availability.

Role is the heaviest term. Efficiency adjusts for quality. Matchup modifies the total within a bounded range. Availability is a gate — it can zero out a player or apply a haircut for injury risk. Percentile rank within position then maps that projected number to 0–100, where a score of 85 means the player ranks in roughly the top 15% of startable players at their position this week.

Four inputs. One number. Nothing hidden.

Why role carries the most weight

Target share is stable in a way that touchdowns and yards-after-catch are not. A receiver who sees 28% of his team's targets in Week 1 will see something close to that in Week 6. The player who got lucky on a 60-yard catch in the rain is harder to project.

Take a receiver like CeeDee Lamb. Even in weeks when the ball didn't fall his way, Dallas's scheme funneled targets toward him. The volume was predictable. The yardage wasn't. The Start Score cares more about the volume.

For receivers, the Start Score uses WOPR — weighted opportunity rate, a published nflverse metric that combines target share and air yards share. Air yards capture opportunity quality, not just volume. A receiver running 10 deep routes but catching two of them still saw real usage. For running backs, it's total opportunities (carries plus targets) combined with snap share. For quarterbacks, pass attempts plus designed rush attempts.

All of it is decay-weighted over the last five games. The most recent game counts for 40% of the weight; the fifth gets 10%. If a receiver just broke into a 30% target share role, that registers fast — you're not averaging it into a six-week fog.

Efficiency: why regression matters more than people think

A receiver who turns eight targets into 32 PPR points is doing something real. A receiver who turns one target into 20 points on a single long catch was probably just in the right place.

Efficiency is PPR points per opportunity, regressed 50% toward the positional mean. That regression pulls outliers toward earth without erasing genuine signal. Think of the boom-or-bust WR who goes nuclear on one deep ball in Week 3, then logs four targets for the next three weeks. Without regression, that one game inflates his efficiency number for weeks. With it, the score treats it for what it was: one data point.

A receiver with high volume and strong conversion efficiency ranks near the top. A big-play WR with a 12% target share ranks lower than his highlight reel suggests, even after one huge game. The formula is not impressed by highlights.

Matchup: it matters less than people say

The matchup component combines two signals: how many fantasy points the opposing defense has allowed to that position this season, and the game's Vegas implied team total. Higher implied totals mean more expected scoring, which creates more opportunity for everyone in that offense.

The multiplier is bounded between 0.85 and 1.15. That cap is deliberate.

People push back on this. A 15% swing either way doesn't feel like much when you're deciding between two borderline starters on a Sunday morning. But consider the alternative: if the matchup term could move a score by 30 or 40 points, you'd end up streaming WR3s into soft defenses every week and ignoring usage entirely. We've seen that strategy lose leagues in real time. The cap isn't arbitrary — it reflects how much matchup actually moves the needle relative to role, which is: a little.

A receiver with a 9% target share facing a terrible defense is still a 9% target share receiver.

Availability: a gate, not a nudge

Injury status isn't a slider. Out, Doubtful, or IR gets a zero. A Questionable player gets a 10% haircut on the full projection, because most questionable players do play, but some don't — and a uniform haircut is more honest than pretending we can predict which Q designations are real.

The score takes the more conservative of nflverse's official injury report and Sleeper's current injury status. If one source says active and the other says questionable, the player gets the questionable multiplier. Better to be slightly conservative on a real starter than confidently wrong on an injury scratch.

The cold start problem: Week 1 is genuinely hard

Before the season starts, there are no current-season games. Week 1 scores run entirely on prior-season usage, and we label them that way.

This creates real uncertainty. Players who changed teams, rookies without NFL snap history, veterans recovering from late-season injuries — the model carries meaningful risk on all of them, and no clever weighting fixes that. We show a lower-confidence label for these players rather than hiding the limitation behind a confident-looking number. By Week 5, the prior-season window has decayed out and the rating is running on live current-season data.

Why simpler beats smarter here

Machine learning models trained on historical data would probably outperform our formula on raw backtested accuracy. We chose not to build one, for one reason: you can't check an ML model.

A neural network trained on five years of nflverse data might be better at predicting certain edge cases. It would be impossible to explain. When it's wrong, you'd get a shrug and a version number. The Start Score is built on a published formula with documented weights. When the number is wrong — and it will be wrong — you can trace exactly why: which input moved, which weight drove it, which assumption didn't hold. That's not a consolation prize. That traceability is the product.

Check this week's ratings, component scores, and projected points position by position at RankFantasy's Start Score.